4 ms·
That'd be because it's all CPU-based at the moment . So your 4080 is taking a vacation while your CPU hits the gym. That's obviously not ideal, but you'll have
by Keavon 2y ago
That'd be because it's all CPU-based at the moment . So your 4080 is taking a vacation while your CPU hits the gym. That's obviously not ideal, but you'll have to trust me that it is due to long-term architectural planning reasons and not a blatant disregard for sensible development practices. Our node graph engine is really advanced—it's actually a scripting language built upon Rust and its type system—which will have some very sizable benefits once it's done being built. But right now, it means vital things like GPU compute have been blocked by a towering pile of other engineering work. But we've nearly completed those prerequisites and should be able to unlock GPU compute in the early parts of next year (this is also blocked on Firefox and Safari shipping WebGPU support, required to use compute shaders in the browser— and on us having the time to support Windows, Mac, and Linux builds via [Tauri](https://tauri.app/ https://tauri.app/)).
In short, please be patient :) The app's architecture is designed with performance that'll make your CPU and GPU scream, but it's a big job building all of it. Especially for raster imagery, that's where a CPU-bottlenecked render pipeline is especially affected. But next year is the year we move on from vector graphics to raster once the GPU is utilized in the render pipeline.
Thanks for taking a look!
- john01dav 2y agoWhen you start supporting GPUs, what APIs do you plan to use (cuda, vulkan, dx, etc.)? It would be quite unfortunate to use a non-vendor-neutral API (for both OS and GPU vendor). I would probably use wgpu or vulkan for this.
- Keavon 2y agoEverything will be using compute shaders for the foreseeable future. [WGPU](https://wgpu.rs/ https://wgpu.rs/) abstracts that to work with WebGPU on browsers, DirectX/Vulkan on Windows, Metal on Mac, and Vulkan on Linux and Android. There may be opportunities to explore vendor-specific options like CUDA in the far future to leverage a further increase in performance, but compute shaders are portable and nearly as good.